Geometry-Based Optimized Point Cloud Compression Methodology for Construction and Infrastructure Management

Laser scanning has been widely used in as-built 3D modeling for construction and infrastructure management. When laser scanners can produce dense point clouds that capture cm-level features in minutes, efficient 3D point cloud processing, storage, and visualization remain challenging due to gigabytes of 3D imagery data. A practical solution is to compress the point cloud while keeping enough geometric information (e.g., edges, corners). Existing compression method uses uniform sampling that reduces data densities everywhere without considering that geometric complexities of different parts of a scene may deserve varying data densities. However, limited studies quantitively assess the deviations between pre-and post-compression point clouds. This research examines a laser-scanning data compression method that enables automatic compression of a point cloud with varying subsampling rates per geometric complexities of parts of data and a user-specified compression ratio (the ratio of the removed point cloud to the original point cloud, the segment with the highest value is the most compressed). Compared with qualitative methods, this method quantifies the relationship between conserved geometric details, subsampling rates, and geometric complexities. First, the developed approach takes segmented point clouds as inputs (either manual or automatic segmentation results) that contain data segments of significantly different geometric complexities depending on the shapes underlying the data segments. Next, surface smoothness and curvature are calculated to quantify the geometric complexities of point cloud segments. Then per total compression ratio, the developed approach assigns various sub-compression ratios to the segments per their geometry complexities (e.g., complicated segments are assigned higher compression ratio to keep more data and vice versa). Finally, this approach evaluates compression results by accessing deviations between the original and compressed point clouds. Results from a case study indicate that the proposed method can achieve a high compression ratio and conserve more information than the uniform sampling approach.